Damasukasu Japanese chef's knife sets are fake

What looks like a great deal on Japanese chef's knives is popping up on various web sites (e.g. Salon, The Daily Beast, Groupon).  Unfortunately the knives aren't at all like what's advertised.

It's a set of three handmade carbon-steel chef's knives from Japan.  The price is great:  regularly $599.99, on sale (one day only) for $49.99!  

Here's the accompanying description:

I've been looking for reasonably priced knives made of carbon steel (not stainless steel), and these sizes are just right, so I bought them with a One-Day Sale offer from the Daily Beast.

But now they're in my kitchen I can see that:

1. They're made of stainless steel, not carbon steel.  I wanted carbon steel for its fine edge and easy sharpening; I don't mind that it stains and rusts when treated badly.

2.  They're machine made, not hand-made.

3.  They don't contain 67 (or any) layers of steel.  Blades made by hammering together many layers of steel have characteristic rippling patterns resulting from minor random variations in the thicknesses of the layers.  But these blades are single solid pieces of stainless steel, with the rippled patterns faked by etching rows of tiny dots in the surface (probably with a laser).  Each blade has exactly the same pattern on both surfaces, and, as can be seen in the photo below, the three knives all have exactly the same pattern (scaled up or down to fit the different blade sizes).

4.  The cutting edges show tiny ridges, suggesting that they were laser-cut, not honed and polished

5.  The handles are resin, not wood.

6.  The box is not cherry oak but cheap wood covered with woodgrain vinyl.

I can't tell where they were made; the only identifying mark is the logo stamped onto the knives and the box. 

Don't get me wrong - these aren't bad knives.  They seem to be very well made and feel good in the hand (well balanced and not heavy).  I think they're excellent value for $50.  But they're certainly not as advertised.

New RNA-seq samples (quality)

The new RNA-seq data arrived about a week and a half ago. Since then I've aligned it to the KW20 genome and looked a bit into the antitoxin knockout samples.

But I wanted to mention the quality of the data. This analysis was done partially using FASTQC:

These plots show the distribution of quality at each base position. (Phred score on y axis, position in read on x axis) In Illumina data, we expect the quality to decrease the longer the read gets. In the old data, you can see the quick drop of quality at the end of each read. These plots are very representative of both old and new datasets and overall I'd say the sequencing quality is higher for the new dataset. This doesn't say anything about the content of the samples though.

Here are some interesting numbers:

Average # of reads of older samples: 7,643,000

Average # of reads of new samples: 11,620,000

On average, the new set of samples are much larger. But there are a few samples that are on the low side:

- toxx_M3_K ( 2,311,395 reads )

- hfqx_M2_J ( 5,416,787 reads )

- hfqx_M3_J ( 2,285,198 reads )

Average % GC in older samples: 40.6%
Average % GC in new samples: 48.8 %

This seems a little alarming at first. Let's look at the distribution more closely: 

This plot shows the number of reads with a given GC content (red line) and a smoothed theoretical bell-curve (blue line). I added the green bars to help compare peaks in the plots. The old data is what I expect a good distribution of KW20 mRNA to look like. The large peaks in the new dataset lines up with tiny bumps in the old data. These are likely from ribosomal RNA. The alignment data also supports the idea that there is much more KW20 rRNA in the new dataset (probably because we used suboptimal amounts of RiboZero when extracting the RNA). Is this a problem? I'll get to that soon.

% Aligned to KW20 genome in older samples (average):  98.1%

% Aligned to KW20 genome in new samples (average): 81.1%

It seems like there is a lot more contamination in the new datasets. In fact, ~20% of reads do not appear to come from KW20. I blasted a few of these unaligned pieces and they all aligned to 23S ribosomal RNA in other bacterial species (E. coli, Bacillus subtilis, etc.). This is a shame, but again, is it a problem?

Average count data in older samples: 8,223,000
Average count data in new samples: 3,000,000

I should explain what this number means. This is a count of the number of reads that align to some gene in the KW20 genome (other than rRNA genes), averaged across all samples. (note: reads that overlap two genes may be double counted). In other words, this number is proportional to the number of useful reads (useful for differential expression).

So there are about 2.75x more useful reads in older samples than the new samples. Although this isn't optimal, I don't think it is a big problem (and I haven't run into an obvious problem yet). Perhaps a few samples could use a couple million more reads, overall the data looks usable.

Recent Lab Meeting

I've uploaded the powerpoint I used during the last lab meeting I did. I'll get around to making a proper post about it eventually, but until then I will have the plots available here:

https://www.slideshare.net/secret/3Eg3j4EsbhTQqS

murE Crystal structures

I've been investigating murE hypercompetence today and I just wanted to make a post for future reference if I ever need it.

I've been looking at crystal structures of murE from here:
http://www.ncbi.nlm.nih.gov/structure/?term=%22mure%22


There are 2 structures from M. tuberculosis. The first shows murE bound to UAG and magnesium. You can see that the ADP binding pocket between the brown and blue subdomains is open.

The second M. tuberculosis murE structure includes ADP:

In E. coli, it's clear that the structure is very similar:

There's also one for Staphylococcus aureus that is again, very similar.

But where do we find the murE point mutations? Looking at the M. tuberculosis structure, we find them here (based on a BLAST alignment):

murE751 is a L -> S substitution found on an alpha helix (the backside where the arrow is pointing).

It looks pretty close to the active site, but it points away from all the action. The other murE point mutations (murE749, G -> R; murE750, G -> W) appear on the same brown domain but far away on the surface of the protein:

The first mutation mentioned may hinder the activity since it is relatively close to the ADP binding area but for the most part, I'm skeptical that any of these point mutations have a big impact on activity. I looked at the location of the point mutations for all three species (M. tuberculosis, E. coli and S. aureus) and it seems very consistent.

Does transcription impact recombination?

The short answer seems to be a no.

So I got Josh's recombination data where he gave Haemophilus influenzae 86-028NP donor DNA to KW20 and sequenced 10,000 pooled colonies that had undergone recombination.

I wanted to see if transcription affected recombination (ex. areas of high recombination tend to recombine less) so I made some plots that combined both sets of data:

Here I graphed the % donor alleles as the points. These points correspond to how much recombination is happening. I added some rectangles (genes) at the bottom to orient the graph. The black line corresponds to transcription. The line is the log of the coverage from KW20 at timepoint M0; This is straight from raw read counts. Note there is no axis for the values, but it is plotted on the same scale between graphs. I just wanted to see relative levels of transcription.

I made plots like these for the whole genome:

I went though all of these plots and there does not seem to be any relationship between transcription and recombination at all. If there is an effect of transcription, it is very small and probably not worth looking into deeper at this point.

If interested, I uploaded all 182 plots and the R script I made for this in the Google Drive:
RNAseq docs/Scott/Recombination

Phase variable genes are a pain

Since I've started looking at the RNA-seq data, there are a few genes that periodically show up as very differentially expressed. Trying to figure out what is going on has been rough, but I think I finally understand. These genes appear to be phase variable.

HI1537 (licA), HI1538 (licB), HI1539 (licC), HI1540 (licD)
has CAAT repeats:

Appears to be strongly downregulated in crpx only

in crpx:

in kw20 (and what seems to be everything other than crpx):

Note: that black segments corresponds to a 4bp deletion. There appears to be an additional repeat in the crpx strain that decreases the transcript abundance.

 HI1287 (hsdM), HI1286 (hsdS), HI1285 (hsdR)

has AGCAG repeats:

Appears to be strongly upregulated in taxx only

in taxx (note the 5 bp deletion):

in kw20 (and what seems to be everything other than taxx):

It looks like the extra repeat in all the strain other than taxx causes a huge difference.

 HI0354 (lic3A)

has AACT repeats:

Note: these repeats go on for more than 100 basepairs. Since we use reads of length 101 I don't think it's possible to detect a deletion here. This is the downside of using short reads.

But this gene is highly upregulated in KW20 in sBHI

Compare kw20:

versus murE at the same timepoint:

I have a good reason that phase variation is also responsible for this difference.

And finally: 

 HI1457 (opa), HI1456 (??)

has no repeats.

This gene pair has been a pain in my side for a while. Mainly because it looks like it is turned on in sBHI only in cells that are hypercompetent:

For strains in BHI, there is like a... 1.6% chance of seeing this assuming each strain has a 50% chance of having the gene on. But in MIV, it looks like crpx (and maybe hfqx) are the only ones that don't really express this gene. There is not a clear view of what's going on in this data.

I dug around some papers and found this one: 

The phasevarion: a genetic system controlling coordinated, random switching of expression of multiple genes.

This paper is oddly pertinent to this blog post. The paper shows that the opacity protein opa is regulated by a type III restriction-modification system. HI1058/HI1059 encodes the mod gene that has tetranucleotide (AGTC) repeats. The number of repeats determine the reading frame. Two reading frames produce protein (72 or 86 kDa) and one doesn't. The paper shows that opa expression is reduced under one of these reading frames. There is likely some methylation happening that blocks opa expression.

Unfortunately, I am not able to see which samples have active mod because again, the repeat region is greater than 100 bp. The mod gene does seem to be expressed (significantly) more in the crp knockout though. For now, I think I'll treat opa as indirectly phase variable.

Things worth mentioning about competence genes in competence mutants

I've turned my focus on trying to figure out how/why the various competence mutants we have alter regular competence. To do this, I'm comparing expression of competence genes between strains.

A quick message about normalization: 
DESeq2 (the R package I used to normalize the count data) normalizes data to allow, for instance, a comparison between gene A in condition X and gene A in condition Y. It does not normalize in a way that makes comparisons between gene A and gene B possible because it does not take into account differences in gene length (longer genes are expected to have more reads).

To make comparisons between genes possible, instead of asking "how much expression of gene A is there?" I ask "how much expression of gene A is there compared to gene A expression in KW20 M0?" I did this by dividing normalized count values by the normalized count in KW20 M0. In essence, KW20 M0 is treated as a baseline and measurements are deviations from that baseline.

Here is some data (error bars are standard error):

The first set of plots shows KW20 competence gene expression. As you can see, most competence genes are ~10-100x induced in M2 relative to M0. Compare this to the sxy or crp knockout where very little induction is seen. Interestingly, HI0365 seems to be particularly down in these knockouts.

The toxin and double toxin/antitoxin knockout strains behave similarly to KW20 (except for the toxin/antitoxin genes, of course) and peak at pretty much the same levels in M2 as KW20. Competence expression in the double knockout (taxx) seems to be a bit lower than KW20 in M3 though. Overall, nothing unexpected.

Knocking out hfq appears to cause slower induction of competence (compare M1 to data above) and definitely does not lead to full induction of the competence regulon when comparing to KW20 M2. Presumably, this failure to completely turn on the competence regulon leads to the 10x comptence defect seen in this strain.

This one is confounding. There's definitely high expression of the toxin/antitoxin pair when the antitoxin is knocked out (at all timepoints!), suggesting that the toxin is self-promoting. This behaviour is seen in other toxin/antitoxin pairs as well. This strain was shown to not be able to uptake DNA, but competence gene expression is comparable to the hfq knockout strain (which is only mildly less competent). The only competence gene that is consistently down in this strain is HI1631 which pretty much nothing is known about (except that it may have some restriction enzyme like function... at least it has a motif that suggests that). Interesting to note though, the 3-enzyme restriction modification system hsdR, hsdM, and hsdS is hugely upregulated in the toxin/antoxin double knockout. This is odd because you'd think removing both the toxin and antitoxin should have no effect whatsoever on the transcriptome (or that you would see the same thing if you just removed the toxin...?)

Note, these two have BHI samples only. Strikingly, rpoD and sxy-1 behave very similarly. Sxy-1 is hypercompetent because of the weakened 5' stem mRNA structure which leads to more Sxy protein and expression of competence genes. Perhaps the rpoD point mutation works in a similar way to increase sxy translatability.

Finally, we have the infamous murE point mutation. Competence gene expression increases a bit from B1 to B2. Apparently competence genes are expressed more in murE in sBHI than KW20 in MIV. At the very least, it's possible to say that murE is hypercompetent because something is causing aggressive expression of competence genes (and not some change of membrane permeability, for example).